Remote

You run a remote MCP server that exposes tools over HTTP or local stdio. This setup lets you connect to your server from MCP clients and execute tools remotely, enabling automation and centralized capabilities without embedding everything directly in a single application.

How to use

Connect to your remote MCP server from an MCP client to access your tools. You can use an HTTP URL to reach a remote server, or you can run a local stdio proxy that forwards commands to the remote or localhost endpoint. Once connected, you can perform tasks by invoking the available tools just like you would with any built-in MCP tool set.

How to install

Prerequisites you need before setting up the MCP server and clients:

• Node.js and npm installed (recommended)
• A Cloudflare account if you plan to deploy via Cloudflare Workers
• Basic familiarity with command line usage

Step-by-step setup to create and run your remote MCP server when you want to deploy to Cloudflare Workers or run locally:

• Install the project scaffolding to create your MCP server
• Define your tools inside the init() method of your source entry point using this.server.tool(...)
• Run the server using the appropriate command for your deployment method (HTTP remote or stdio local proxy)

Commands shown for deployment and local testing include the following examples. Use what matches your environment and desired deployment target.

Additional content

Customize your server by adding tools inside the init() function of your MCP server entry point. Each tool is registered with this.server.tool(...) to extend the MCP with new capabilities that you can invoke from clients.

Connect to Cloudflare AI Playground to access your MCP server as a remote client. Open the playground, enter your deployed server URL, and start using your tools directly from the playground interface.

If you prefer running a local client, you can connect Claude Desktop to your remote MCP server by configuring the mcp-remote proxy. In Claude Desktop, go to Settings > Developer > Edit Config and provide the following config to expose your tools via an HTTP endpoint.